基于多分类SVM和Hd的目标跟踪算法  

Tracking algorithm based on MultiClass SVM and Hausdorff distance

作　　者：苗超维 秦品乐[1] MIAO Chao-wei QIN Pin-le(College of Computer and Control Engineering, North University of China, Taiyuan 030051, China)

机构地区：[1]中北大学计算机与控制学院

出　　处：《计算机工程与设计》2016年第11期3118-3123,共6页Computer Engineering and Design

摘　　要：针对传统基于支持向量机的目标跟踪算法中计算复杂度高,目标由于严重遮挡或者出离场景导致的目标跟踪漂移,提出一种基于结构化多分类SVM和Hausdorff距离的目标跟踪算法。通过提取相邻帧之间的canny特征算子,计算目标轮廓特征点的Hausdorff距离,整合相邻帧的图像跟踪序列,对样本学习的采集进行预判,避免传统算法中不必要的样本在线学习;采用结构化多分类SVM目标输出预测函数增加目标变换种类,增强目标跟踪的鲁棒性和准确性。实验结果表明,该算法延续了支持向量机良好的泛化能力,可以有效跟踪目标的各种变换,较传统方法有更好的鲁棒性和计算效率。Computational complexity of traditional target tracking algorithm based on support vector machine（SVM）is high and is too excessively robust for target transformation which causing drifting,to solve the problems,a target tracking algorithm based on structured output MultiClass SVM and Hausdorff distance was put forward.Through extracting the characteristics of canny operator between adjacent frames,the target contour feature points of Hausdorff distance were calculated,and the adjacent frame sequences of tracked image were integrated.Samples of learning were anticipated,avoiding the unnecessary samples in traditional algorithm online learning.The structured output target prediction function was used to increase additional categories of target transformation,greatly enhancing the robustness and accuracy of target tracking.Experimental results show that the algorithm not only maintains the generalization ability of the support vector machine（SVM）,but effectively tracks the target with various transformations,which shows better robustness and computational efficiency than the traditional methods.

关 键 词：多分类SVM HAUSDORFF距离 目标跟踪 支持向量机 canny特征算子 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]